Underground water exploitable quantity evaluation method and system based on coupling of physical mechanism and neural network

Abstract

The invention discloses an underground water exploitable quantity evaluation method and system based on coupling of a physical mechanism and a neural network, and relates to the technical field of water resource management, meteorological data, surface water hydrological monitoring data, geological survey data, underground water resource development and utilization data and hydrogeological parameter data are integrated and preprocessed, and a basic data set is obtained; a three-dimensional geological model of a heterogeneous structure is constructed by utilizing a digital twin physical engine and combining drilling geological data. By introducing a digital twin technology and a numerical model, the underground water flowing process can be simulated more accurately, a digital twin physical engine is utilized to construct a heterogeneous aquifer digital model, a dynamic interaction process of surface water and underground water is embedded, and the model is enhanced through a graph neural network and a physical information neural network, so that the underground water flowing process is simulated more accurately. Errors caused by hypothesis of aquifer homogeneity, neglecting of surface water-underground water coupling process and the like of a traditional model are effectively reduced, and the accuracy of underground water exploitable quantity evaluation is improved.

Description

Technical Field

[0001] The present invention relates to the technical field of water resource management, and in particular to a method and system for evaluating groundwater exploitability based on physical mechanism and neural network coupling. Background Art

[0002] With population growth, accelerated urbanization and agricultural industrialization, the demand for water resources continues to rise. Long-term over-exploitation will lead to a drop in groundwater levels. In severe cases, it will also cause geological disasters such as ground subsidence, ground fissures, and seawater intrusion, threatening the safety of infrastructure. Water resource carrying capacity refers to the maximum scale of social and economic development supported by the water resource system within a specific time and space scope while maintaining the ecological and environmental functions. By quantifying the amount of exploitable groundwater and clarifying the upper limit of regional water resource development, we can avoid the extensive model of "supply determined by demand" and shift to sustainable management of "supply determined by demand".

[0003] In the existing technology, hydrogeological models are over-simplified. Traditional MODFLOW numerical models often assume that aquifers are homogeneous, ignoring the existing heterogeneous structures, and neglecting the coupling process of surface water-groundwater interaction and evapotranspiration. There are certain simulation errors, which interfere with the evaluation of groundwater exploitation. Therefore, how to construct a digital model of heterogeneous aquifers, embed the dynamic interaction process between surface water and groundwater, and use graph neural networks to replace traditional numerical solvers, embed Darcy's law and mass conservation constraints into model training through physical information neural networks, and use Bayesian layers to quantify parameter uncertainties to ensure accurate evaluation of groundwater exploitation. This is the problem to be solved by the present invention. To this end, a groundwater exploitation evaluation method and system based on the coupling of physical mechanisms and neural networks are proposed. Summary of the Invention

[0004] The purpose of the present invention is to provide a method and system for evaluating groundwater recoverability based on physical mechanism and neural network coupling to solve the problems raised in the above background technology.

[0005] In order to solve the above technical problems, the technical solution adopted by the present invention is: First, a method for evaluating groundwater recoverability based on the coupling of physical mechanisms and neural networks includes the following steps: S1. Integrate meteorological data, surface water hydrological monitoring data, geological survey data, groundwater resource development and utilization data, and hydrogeological parameter data, and pre-process them to obtain a basic data set; S2. Using the digital twin physics engine and combining it with borehole geological data, we construct a three-dimensional geological model of the heterogeneous structure, characterize the heterogeneous structure, and restore the true spatial distribution characteristics of the aquifer. S3. Embed the dynamic interaction process between surface water and groundwater into the 3D geological model, simulate the coupling relationship between the two, and improve the water cycle simulation system; S4. Introducing graph neural networks to replace traditional numerical solvers in groundwater flow systems, embedding Darcy's law and mass conservation equations into the loss function, and strengthening the physical consistency of the model through physical information neural networks; S5. Introduce a Bayesian neural network layer to model the probability distribution of permeability coefficient, water supply degree or water release coefficient, and quantify the impact of input data errors on the uncertainty of simulation results; S6. Build an interactive digital twin platform to display the amount of groundwater that can be extracted in real time, and support managers to dynamically revise extraction strategies and carrying capacity red lines based on monitoring data.

[0006] A further improvement of the technical solution of the present invention is that: S1 specifically includes: Collect meteorological data, surface water hydrological monitoring data, geological survey data, groundwater resource development and utilization data, and hydrogeological parameter data from various sources, and compile a list of data sources to ensure data integrity and accuracy; Clean the collected data to remove duplicates, anomalies, and missing values, and standardize data from different sources and formats to unify the data units and formats. Specifically, standardize the time intervals of time series data to improve data quality through data cleaning and standardization. Integrate the cleaned and standardized data into a unified data platform to build a basic data set, classify and associate the data according to geographic spatial location and time series to ensure the logic and consistency of the data. The basic data set includes complete meteorological data, surface water hydrological monitoring data, geological survey data, groundwater resource development and utilization data, and hydrogeological parameter data.

[0007] A further improvement of the technical solution of the present invention is that: S2 specifically includes: Prepare the digital twin physics engine (OpenGeoSys). Extract cleaned and standardized borehole geological data from the basic dataset and import it into the digital twin physics engine. This includes data such as borehole records, lithology distribution, and permeability coefficient. The data is then spatially positioned and coordinate verified to ensure that the coordinates of the imported data conform to the geographic coordinate system. Elevation verification is performed to ensure that the elevation data is consistent with the actual terrain elevation. Within the digital twin physics engine, visualize the data points as a 3D point cloud to verify that their spatial distribution is reasonable and ensure that the coordinates of each data point are accurate. Using the modeling tools of the digital twin physics engine, combined with borehole geological data, a 3D geological model was constructed. A 3D grid was generated based on the scope and resolution of the study area. Based on the borehole logs and lithologic distribution data, different geological layers and aquifers were divided. Kriging was used to generate a continuous 3D spatial distribution of the aquifers. In the digital twin physics engine, select "Interpolation > Kriging," set the interpolation parameters for the search range, and interpolate the permeability and lithologic parameters to generate a continuous 3D distribution. The heterogeneous structure of the aquifer, including pores, cracks, solution gaps and lithologic changes, is depicted in the 3D geological model. At the same time, the real spatial distribution characteristics of the aquifer are restored based on geological profiles and drilling data.

[0008] A further improvement of the technical solution of the present invention is that: S3 specifically includes: The spatial location and boundary range data of surface water bodies are obtained from hydrological maps, remote sensing image interpretation, and geographic information system databases. The spatial location and boundary range data of surface water bodies are then imported into the three-dimensional geological model. Based on the surface water monitoring data, the initial water level and flow rate of the surface water bodies are set. The initial water level can be obtained from the data of the water level monitoring station, and the initial flow rate can be obtained from the data of the flow monitoring station. Then, combined with the rainfall data, the recharge sources of precipitation and surface water are defined. The rainfall data can be obtained from the observation data of the meteorological station. The dynamic interaction between surface water and groundwater is embedded in the 3D geological model. The permeability coefficient and hydraulic conductivity coefficient are set based on the borehole geological data. The recharge path and rate of surface water to groundwater, as well as the discharge process of groundwater to surface water, are defined. Darcy's law and the mass conservation equation are used to simulate the coupling relationship between surface water and groundwater, ensuring that the model can accurately reflect the water exchange between surface water and groundwater. A rainfall module is added to the model to simulate the recharge effect of rainfall on surface water and groundwater, and to distribute rainfall to the corresponding areas of surface water bodies and aquifers. At the same time, the impact of evaporation and transpiration on groundwater levels and surface water volume is analyzed, and the rainfall, evaporation and transpiration processes are integrated to improve the water cycle simulation system.

[0009] A further improvement of the technical solution of the present invention is that: S4 specifically includes: The groundwater flow system is represented as a graph structure, in which nodes represent key locations in the aquifer, such as monitoring points and borehole locations. Each node includes attributes such as coordinates, hydraulic head, lithology, and permeability coefficient. The coordinates represent the position of the node in three-dimensional space (longitude, latitude, and elevation), the hydraulic head represents the groundwater level at the node, the lithology represents the lithology type at the node, and the permeability coefficient represents the permeability coefficient at the node. The edge represents the hydraulic connection between the nodes. Each edge includes attributes such as the connecting node, distance, and hydraulic conductivity coefficient. The connecting node represents the two nodes connected by the edge, and the distance represents the Euclidean distance between the two nodes. The hydraulic conductivity coefficient is calculated based on the permeability coefficient and distance. A graph neural network is constructed based on this graph structure, and the network architecture is designed to adapt to the complexity of groundwater flow. The graph neural network includes an input layer, a graph convolution layer, a hidden layer, and an output layer. Darcy's law and the mass conservation equation are embedded in the loss function of the graph neural network. Darcy's law is used to describe the relationship between the flow rate of groundwater in the aquifer and the head gradient. The mass conservation equation is used to ensure that the water inflow is equal to the outflow plus the storage variable. This enables the graph neural network to automatically learn solutions that satisfy physical laws during training, thereby improving the physical consistency and prediction accuracy of the model. The physical information neural network (PINN) technology is used to enhance the training process of the graph neural network model. The physical information neural network simulates the physical process of groundwater flow by directly embedding physical laws into the training objectives of the network.

[0010] A further improvement of the technical solution of the present invention is that the training process of the graph neural network model is: Embed Darcy's law and the mass conservation equation into the loss function of the graph neural network. The loss function of the graph neural network consists of two parts: data-driven loss and physical constraint loss. The data-driven loss minimizes the mean square error between the predicted head and the actual monitored head. The physical constraint loss covers Darcy's law and the mass conservation equation. Darcy's law is used to describe the relationship between groundwater flow rate and head gradient. The mass conservation equation ensures that the water inflow is equal to the outflow plus the storage variable. Embedding Darcy's law and the mass conservation equation into the loss function results in the physical constraint loss. Using Physically Informed Neural Network (PINN) technology, physical laws are directly embedded into the network's training objectives. During training, data-driven solutions are learned while satisfying physical laws. Gradient descent is then used to optimize network parameters, satisfying both data-driven loss and physical constraint loss. Prepare training data and validation data to train the graph neural network model. The training data includes the initial water head, lithology, permeability coefficient, etc. of the nodes, and use the gradient descent method to optimize the network parameters and minimize the total loss function. In each training step, calculate the data-driven loss and physical constraint loss, update the network parameters, and then evaluate the model performance and check whether the model satisfies the physical laws. Adjust the network architecture and hyperparameters based on the verification results, use the trained graph neural network model to predict groundwater flow, analyze the model output, and evaluate the dynamic changes of groundwater flow.

[0011] A further improvement of the technical solution of the present invention is that: S5 specifically includes: A Bayesian neural network layer is introduced into the graph neural network (GNN) architecture to perform probability distribution modeling on the permeability coefficient, water supply degree, or water release coefficient. The permeability coefficient, water supply degree, or water release coefficient is modeled as a combination of mean and variance. The mean represents the expected value of the parameter, and the variance represents the uncertainty of the parameter. The Bayesian layer represents the uncertainty of the parameter by introducing a probability distribution (Gaussian distribution). The Bayesian neural network layer is used to quantify the input data error. The Monte Carlo sampling method is used to simulate groundwater flow under different input data error scenarios and calculate the corresponding simulation results. Through multiple sampling, the distribution characteristics of the simulation results, including mean, variance and confidence interval, are analyzed to quantify the uncertainty impact of input data error on the simulation results. Analyze and evaluate the uncertainty of simulation results. By comparing simulation results under different input data error scenarios, identify the sources of uncertainty, including parameter uncertainty, data uncertainty, and model uncertainty, and evaluate the impact of uncertainty on decision support.

[0012] A further improvement of the technical solution of the present invention is that the process of quantizing the input data error is: Select the Monte Carlo sampling method and set the number of samples M to randomly draw samples from the defined probability distribution. The number of samples should be large enough to ensure that the uncertainty of the parameters can be fully captured. For each sampling, a set of values ​​are randomly drawn from the probability distribution of the permeability coefficient, water supply degree or water release coefficient, and the groundwater flow under different input data error scenarios is simulated. Each sampling generates a set of permeability coefficient, water supply degree or water release coefficient values, runs the groundwater flow model, and calculates the corresponding simulation results, wherein the permeability coefficient, water supply degree or water release coefficient obtained for each sampling is used as the model input parameter, and the groundwater flow model is run using the model input parameters to calculate the simulation results, which are the groundwater level. Then, the simulation results of each sampling are recorded to form a data set containing M simulation results; The simulation results obtained from multiple samplings are statistically analyzed, and their distribution characteristics, including mean, variance and confidence interval, are calculated to quantify the uncertainty impact of input data errors on the simulation results.

[0013] A further improvement of the technical solution of the present invention is that: S6 specifically includes: Build an interactive digital twin platform that integrates groundwater monitoring data, model prediction data, and water resource management information. It also provides real-time data access and supports access to multiple data sources. At the same time, using geographic information system (GIS) technology, it visualizes the amount of groundwater available for extraction on a map. The interactive digital twin platform receives groundwater monitoring data in real time, including key indicators such as water level and flow rate. It uses this data to dynamically calibrate the groundwater flow model to ensure the accuracy and timeliness of model predictions. When there is a deviation between the groundwater monitoring data and the model prediction, the model correction process is automatically triggered to adjust the model parameters or update the model structure to improve the accuracy of the groundwater recoverable volume prediction. Based on real-time monitoring data and model prediction results, managers are supported to dynamically adjust groundwater extraction strategies. When the extraction volume approaches or exceeds the preset carrying capacity red line, an early warning is automatically issued. The early warning information is notified to managers through the user interface, SMS or email, and optimization suggestions are provided. Then, according to the analysis results and early warning information provided by the interactive digital twin platform, the extraction plan is adjusted to ensure the sustainable use of groundwater.

[0014] In the second aspect, a groundwater exploitation quantity evaluation system based on the coupling of physical mechanism and neural network is used to implement the above-mentioned groundwater exploitation quantity evaluation method based on the coupling of physical mechanism and neural network, including: Data integration and processing module, used to collect, clean, standardize and integrate meteorological data, surface water hydrological monitoring data, geological survey data, groundwater resource development and utilization data and hydrogeological parameter data from different sources to build a unified basic data set; The geological model construction module is used to build a three-dimensional geological model of heterogeneous structures using the digital twin physics engine and borehole geological data, restore the real spatial distribution characteristics of the aquifer, and improve the physical authenticity of the model; The groundwater flow simulation module is used to embed the dynamic interaction process between surface water and groundwater in the 3D geological model, simulate the coupling relationship between the two, and improve the water cycle simulation system; Uncertainty Quantification Module, which introduces graph neural networks to replace traditional numerical solvers in groundwater flow systems, embeds physical laws, and quantifies parameter uncertainties through a Bayesian neural network layer to ensure the accuracy and reliability of simulation results, improving the model's prediction accuracy and physical consistency; The interactive digital twin platform displays the exploitable amount of groundwater in real time by building an interactive digital twin platform, supporting managers to dynamically revise exploitation strategies and carrying capacity red lines based on monitoring data, thus achieving real-time display and dynamic management of the exploitable amount of groundwater.

[0015] Due to the adoption of the above technical solution, the present invention has the following technical advancements compared to the prior art: The present invention provides a method and system for evaluating groundwater exploitation based on the coupling of physical mechanisms and neural networks. By introducing digital twin technology and numerical models, it can more accurately simulate the groundwater flow process, use the digital twin physical engine to construct a digital model of heterogeneous aquifers, embed the dynamic interaction process of surface water and groundwater, and enhance the physical consistency of the model through graph neural networks and physical information neural networks. It can effectively reduce the errors caused by traditional models due to the assumption of aquifer homogeneity and the neglect of surface water-groundwater coupling processes, thereby significantly improving the accuracy of groundwater exploitation evaluation.

[0016] The present invention provides a groundwater exploitation evaluation method and system based on the coupling of physical mechanisms and neural networks. By introducing a Bayesian neural network layer, a probability distribution model is performed on the permeability coefficient, water yield or water release coefficient. The Monte Carlo sampling method is used to quantify the uncertainty impact of input data errors on the simulation results. This method can not only provide a predicted value of the groundwater exploitation volume, but also evaluate the uncertainty range of the predicted results, including the mean, variance and confidence interval. The quantitative analysis of uncertainty provides more comprehensive information, thereby enhancing the scientific nature and reliability of decision-making. BRIEF DESCRIPTION OF THE DRAWINGS

[0017] In order to more clearly illustrate the embodiments of the present application or the technical solutions in the prior art, the following briefly introduces the drawings required for use in the embodiments. Obviously, the drawings described below are only some embodiments described in the present invention. For ordinary technicians in this field, other drawings can also be obtained based on these drawings.

[0018] Figure 1 Schematic diagram of the workflow of the method and system for evaluating groundwater recoverability based on physical mechanism and neural network coupling of the present invention; Figure 2 The figure is a flow chart of the groundwater exploitable quantity evaluation method based on the coupling of physical mechanism and neural network of the present invention. DETAILED DESCRIPTION

[0019] To make the objectives, technical solutions, and advantages of the embodiments of the present invention more clear, the technical solutions in the embodiments of the present invention will be clearly and completely described below in conjunction with the accompanying drawings of the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. All other embodiments obtained by ordinary technicians in this field based on the embodiments of the present invention without making any creative efforts shall fall within the scope of protection of the present invention.

[0020] Example 1, as Figure 1 、 Figure 2 As shown, the present invention provides a method for evaluating groundwater recoverability based on physical mechanism and neural network coupling, comprising the following steps: S1. Integrate meteorological data, surface water hydrological monitoring data, geological survey data, groundwater resource development and utilization data, and hydrogeological parameter data, and perform preprocessing to obtain a basic data set. Collect meteorological data, surface water hydrological monitoring data, geological survey data, groundwater resource development and utilization data, and hydrogeological parameter data from different sources, and organize a list of data sources to ensure the integrity and accuracy of the data. Among them, geological survey data include drilling records, lithology distribution, and geological profiles. Drilling records cover information such as borehole depth, lithology description, and groundwater level, and are derived from geological exploration reports, drilling logs, and geological survey agencies. Lithology distribution covers the distribution range, thickness, and burial depth of different lithologies, and is derived from geological maps, geological profiles, and field survey records. Geological profiles must indicate information such as the lithology, thickness, and inclination of different strata, and are derived from geological survey reports and drilling geological databases. Hydrogeological parameters include permeability, water supply, and other parameters. Permeability covers permeability data for different lithologies and different regions, and is derived from hydrogeological data. Quality experiment report, pumping test data, water supply degree covers water supply degree data, which reflects the water release capacity of the aquifer and is derived from pumping test and water injection test reports. Other parameters include water storage rate, hydraulic conductivity and other parameters, which are derived from hydrogeological research reports and monitoring data. Surface water data include river water level and flow as well as lake water level and water volume. River water level and flow cover river water level and flow monitoring data, which are derived from hydrological monitoring stations, hydrological yearbooks, and hydrological databases. Lake water level and water volume cover lake water level and water volume change data, which are derived from lake monitoring stations, remote sensing data, and hydrological models. The collected data are cleaned to remove duplicates, anomalies, and missing values, and data from different sources and formats are standardized to unify the data units and formats. Among them, the time interval of time series data is standardized. Through data cleaning and standardization, the data quality is improved. The cleaned and standardized data are integrated into a unified data platform to construct a basic data set. The data are classified and associated according to geographic spatial location and time series to ensure the logic and consistency of the data. S2. Use the digital twin physics engine and combine it with borehole geological data to build a three-dimensional geological model of heterogeneous structures, characterize the heterogeneous structure, restore the real spatial distribution characteristics of the aquifer, prepare the digital twin physics engine (OpenGeoSys), extract the cleaned and standardized borehole geological data from the basic data set and import it into the digital twin physics engine, including drilling records, lithology distribution, permeability coefficient and other data, and spatially locate the data, perform coordinate verification, check whether the coordinates of the imported data conform to the geographic coordinate system, perform elevation verification, and ensure that the elevation data is consistent with the actual terrain elevation. In the digital twin physics engine, visualize the data points in the form of a three-dimensional point cloud, check whether the spatial distribution of the data points is reasonable, and ensure that each data point The coordinates of the borehole data are accurate. The borehole log includes the borehole number, coordinates (longitude, latitude, elevation), lithology, groundwater level, etc. The lithology distribution includes the lithology type and distribution range (polygonal area). The permeability coefficient includes the lithology and permeability coefficient value. The geological profile includes the profile location, lithology layering, groundwater level, etc. The modeling tools of the digital twin physics engine are used to build a 3D geological model in combination with the borehole geological data. According to the scope and resolution of the study area, a 3D grid is generated. Different geological layers and aquifers are divided according to the borehole log and lithology distribution data. The Kriging method is used to generate a continuous 3D spatial distribution of the aquifer. In the digital twin physics engine, select "Interpolation > Kriging is a method to set the interpolation parameters of the search range, interpolate the permeability coefficient and lithology parameters, generate a three-dimensional continuous distribution, and depict the heterogeneous structure of the aquifer in the three-dimensional geological model, including pores, cracks, solution cracks and lithology changes. At the same time, according to the geological profile and drilling data, the real spatial distribution characteristics of the aquifer are restored. Among them, according to the geological survey data, cracks are added to the model. The cracks can be represented by line segments or planes, and the hydraulic conductivity parameters of the cracks are set. According to the geological profile and drilling data, the location and size of the cave are determined. The cave is represented in the model as a three-dimensional entity, and its hydraulic characteristics are set. According to the lithology distribution data, the lithology of different areas in the model is adjusted to ensure the model The model can reflect the spatial changes in lithology, and then the geological profile is compared with the three-dimensional geological model to check whether the model can accurately reflect the lithology stratification and groundwater level changes in the profile. If the model is found to be inconsistent with the profile, the lithology distribution and groundwater level in the model are adjusted, and the lithology, groundwater level and other information in the borehole record are compared with the corresponding positions in the model. If the lithology or groundwater level in the model does not match the borehole record, the relevant parameters in the model are adjusted. The accuracy of the model is verified by comparing it with known borehole geological data (other borehole data, monitoring data). If there is a deviation in the model, the model parameters are further adjusted until the model can truly reflect the spatial distribution characteristics of the aquifer; S3. Embed the dynamic interaction process between surface water and groundwater in the three-dimensional geological model, simulate the coupling relationship between the two, and improve the water cycle simulation system. Obtain the spatial position and boundary range data of surface water from hydrological maps, remote sensing image interpretation, and geographic information system databases, and import the spatial position and boundary range data of surface water into the three-dimensional geological model. Set the initial water level and flow of the surface water according to the surface water monitoring data. The initial water level can be obtained from the data of the water level monitoring station, and the initial flow can be obtained from the data of the flow monitoring station. Then, combine the rainfall data to define the recharge source of precipitation and surface water. The rainfall data can be obtained from the observation data of the meteorological station. Embed the dynamic interaction process between surface water and groundwater in the three-dimensional geological model. Set the permeability coefficient and hydraulic conductivity coefficient according to the borehole geological data, define the recharge path and rate of surface water to groundwater, and the discharge process of groundwater to surface water, and use Darcy's law and mass conservation equation to simulate the coupling relationship between surface water and groundwater to ensure that the model can accurately reflect the water exchange between surface water and groundwater. In the model, the permeability coefficient can be obtained through geological survey data, pumping tests, etc., and the hydraulic conductivity coefficient can be calculated by the permeability coefficient and the thickness of the aquifer. The recharge path is determined according to the boundary of the surface water body and the permeability coefficient of the aquifer. The recharge rate is calculated using Darcy's law. The discharge path is determined according to the relative position of the groundwater level and the surface water level. A rainfall module is added to the model to simulate the recharge effect of rainfall on surface water and groundwater, and the rainfall is distributed to the corresponding areas of the surface water body and the aquifer. At the same time, the effects of evaporation and transpiration on the groundwater level and The impact of surface water volume, integrating rainfall, evaporation, and transpiration processes, improving the water cycle simulation system, and dynamically simulating the entire water cycle process to ensure that the model can fully reflect the dynamic changes of the regional water cycle. Among them, evaporation mainly affects the water volume of surface water bodies, and transpiration mainly affects the groundwater level, especially shallow groundwater. The evaporation and transpiration rates can be estimated based on meteorological data and vegetation cover. Rainfall replenishes surface water and groundwater, while evaporation and transpiration consume surface water and groundwater. Water volume is exchanged between surface water and groundwater through infiltration and drainage. S4. Introduce graph neural network to replace traditional numerical solvers in groundwater flow system, embed Darcy's law and mass conservation equation into loss function, and strengthen the physical consistency of the model through physical information neural network, and represent the groundwater flow system as a graph structure, where nodes represent key locations in the aquifer, such as monitoring points, borehole locations, etc. Each node includes attributes of coordinates, head, lithology and permeability coefficient. Coordinates represent the position of the node in three-dimensional space (longitude, latitude, elevation), head represents the groundwater level at the node, lithology represents the lithology type at the node, permeability coefficient represents the permeability coefficient at the node, and edges represent the hydraulic connection between nodes. Each edge includes attributes of connecting nodes, distance and hydraulic conductivity coefficient. Connecting nodes represent the two nodes connected by the edge, and distance represents the Euclidean distance between the two nodes. The hydraulic conductivity coefficient is calculated based on the permeability coefficient and distance. Graph neural network is constructed based on this graph structure, and the network architecture is designed to adapt to the complexity of groundwater flow. Graph neural network includes input Layer, graph convolution layer, hidden layer and output layer. Among them, the input layer inputs the feature vector of the node, including coordinates, water head, lithology, permeability, etc. The graph convolution layer uses graph convolution operation to update the node features, considering the influence of hydraulic connections (edges) between nodes on the node features. The hidden layer is a multi-layer graph convolution layer, which gradually extracts the complex characteristics of groundwater flow. The output layer outputs the water head prediction value of each node. Darcy's law and mass conservation equation are embedded in the loss function of the graph neural network. Darcy's law is used to describe the relationship between the flow rate of groundwater in the aquifer and the water head gradient. The mass conservation equation ensures that the water inflow is equal to the outflow plus the storage variable, so that the graph neural network can automatically learn solutions that satisfy physical laws during training, thereby improving the physical consistency and prediction accuracy of the model. The physical information neural network (PINN) technology is used to enhance the training process of the graph neural network model. The physical information neural network simulates the physical process of groundwater flow by directly embedding physical laws into the training objectives of the network. In addition, the training process of the graph neural network model is: Darcy's law and the mass conservation equation are embedded in the loss function of the graph neural network. The loss function of the graph neural network includes two parts: data-driven loss and physical constraint loss. The data-driven loss is to minimize the mean square error between the predicted water head and the actual monitored water head. The physical constraint loss covers Darcy's law and the mass conservation equation. Darcy's law is used to describe the relationship between the groundwater flow rate and the head gradient. The mass conservation equation ensures that the inflow of water is equal to the outflow plus the storage variable. Darcy's law and the mass conservation equation are embedded in the loss function to obtain the physical constraint loss. The physical information neural network (PINN) technology is used to directly embed the physical laws into the training objectives of the network. During the training process, the learning Learn data-driven solutions and satisfy the laws of physics, then use the gradient descent method to optimize network parameters while satisfying data-driven loss and physical constraint loss. Prepare training data and validation data to train the graph neural network model. The training data includes the initial water head, lithology, permeability, etc. of the nodes, and use the gradient descent method to optimize network parameters and minimize the total loss function. In each training step, calculate the data-driven loss and physical constraint loss, update the network parameters, and then evaluate the model performance. Check whether the model satisfies the laws of physics. Adjust the network architecture and hyperparameters based on the verification results. Use the trained graph neural network model to predict groundwater flow, analyze the model output, and evaluate the dynamic changes of groundwater flow. The calculation formula of data-driven loss is as follows: ; Where, is the data-driven loss, is the total number of nodes, To predict the hydraulic head, is the actual water head; The calculation formula for physical constraint loss is as follows: ; Where, is the physical constraint loss, and are weight coefficients, which are used to balance data-driven loss and physical constraint loss. For nodes The water velocity vector at For nodes The permeability coefficient at For nodes The head gradient at For nodes The divergence of the water velocity vector at , For nodes The water supply degree or water release coefficient at For nodes The rate of change of water head at with time; The calculation formula of the total loss function is as follows: ; Where, is the total loss function; S5. Introduce a Bayesian neural network layer to model the probability distribution of permeability coefficient, water supply degree or water release coefficient, and quantify the impact of input data errors on the uncertainty of simulation results; S6. Build an interactive digital twin platform to display the amount of groundwater that can be extracted in real time, and support managers to dynamically revise extraction strategies and carrying capacity red lines based on monitoring data.

[0021] Example 2, as Figure 1 、 Figure 2 As shown, based on Example 1, the present invention provides a technical solution: preferably, S5 specifically includes: The Bayesian neural network layer is introduced into the graph neural network (GNN) architecture to perform probability distribution modeling on the permeability coefficient, water supply degree or water release coefficient. The permeability coefficient, water supply degree or water release coefficient is modeled as a combination of mean and variance. The mean represents the expected value of the parameter, and the variance represents the uncertainty of the parameter. The Bayesian layer represents the uncertainty of the parameter by introducing probability distribution (Gaussian distribution). The Bayesian neural network layer is used to quantify the input data error. The Monte Carlo method is used to calculate the probability distribution of the parameter. Carlo sampling method is used to simulate groundwater flow under different input data error scenarios and calculate the corresponding simulation results. Through multiple sampling, the distribution characteristics of the simulation results, including mean, variance, and confidence interval, are analyzed to quantify the impact of input data error on the uncertainty of the simulation results. The uncertainty of the simulation results is analyzed and evaluated. By comparing the simulation results under different input data error scenarios, the sources of uncertainty are identified, including parameter uncertainty, data uncertainty, and model uncertainty. Parameter uncertainty refers to the uncertainty of the permeability coefficient, water supply degree, or water release coefficient. Data uncertainty refers to the error in the input data. Model uncertainty refers to the error caused by model assumptions and simplifications. The impact of uncertainty on decision support is evaluated, such as whether uncertainty affects the assessment of groundwater recoverability and whether uncertainty affects water resources. Furthermore, the process of quantifying the error in the input data is: The Monte Carlo sampling method is selected, and the number of samplings M is set. Samples are randomly drawn from the defined probability distribution. The number of samplings must be large enough to ensure that the uncertainty of the parameters can be fully captured. For each sampling, a set of values ​​are randomly drawn from the probability distribution of the permeability coefficient, water supply degree, or water release coefficient. The groundwater flow under different input data error scenarios is simulated. Each sampling generates a set of permeability coefficient, water supply degree, or water release coefficient values. The groundwater flow model is run and the corresponding simulation results are calculated. The permeability coefficient, water supply degree, or water release coefficient obtained from each sampling is used as the model input parameter. The model input parameters are used to run the groundwater flow model and the simulation results are calculated. The simulation result is the groundwater level, and the simulation result of each sampling is recorded to form a data set containing M simulation results. The simulation results obtained by multiple sampling are statistically analyzed, and their distribution characteristics, including mean, variance and confidence interval, are calculated to quantify the uncertainty impact of input data error on the simulation results. Among them, the mean of the simulation result represents the expected value of the simulation result, reflecting the average behavior of groundwater flow under the condition of parameter uncertainty. The variance represents the degree of dispersion of the simulation result, reflecting the uncertainty impact of input data error on the simulation result. The confidence interval provides a reliable range of the simulation result, reflecting the uncertainty of the simulation result under a certain confidence level. The calculation formula for the mean is as follows: ; Where, is the mean of the simulation results, which represents the expected value of the simulation results. is the number of Monte Carlo sampling, that is, the total number of simulations, For the The simulation results obtained by sampling times, the mean reflects the central trend of the simulation results. As the number of Monte Carlo sampling increases, the mean will gradually stabilize near the true expected value; The calculation formula for variance is as follows: ; Where, The variance of the simulation results indicates the degree of dispersion of the simulation results. The larger the variance, the greater the dispersion of the simulation results. The variance reflects the dispersion of the simulation results. As the number of Monte Carlo sampling increases, the variance will gradually stabilize near the true variance value. The calculation formula for the confidence interval is as follows: ; Where, is the confidence interval of the simulation results, which indicates the reliable range of the simulation results at a certain confidence level. is the Z value corresponding to the confidence level, for a 95% confidence interval, ,The width of the confidence interval is determined by the variance and the confidence level. The confidence interval provides a reliable range of the simulation results. As the number of Monte Carlo sampling increases, the confidence interval will gradually stabilize near the true confidence interval. The width of the confidence interval reflects the uncertainty of the simulation results. The wider the width, the greater the uncertainty. S6 specifically includes: Build an interactive digital twin platform to integrate groundwater monitoring data, model prediction data and water resources management information, and have real-time data access function, support access to multiple data sources, and at the same time, through geographic information system (GIS) technology, display the amount of groundwater exploitable in a visual form on the map. The architecture of the interactive digital twin platform includes the front-end, back-end and database. The front-end: user interface (UI), supports data visualization and interactive operations, the back-end: data processing and model calculation module, supports real-time data access and dynamic calibration of the model, the database: stores groundwater monitoring data, model prediction data and water resources management information, the interactive digital twin platform receives groundwater monitoring data in real time, including water level, flow, The groundwater flow model is dynamically calibrated using groundwater monitoring data based on key indicators such as volume to ensure the accuracy and timeliness of model predictions. When there is a deviation between groundwater monitoring data and model predictions, the model correction process is automatically triggered to adjust model parameters or update model structure to improve the accuracy of groundwater exploitable volume predictions. Based on real-time monitoring data and model prediction results, managers are supported to dynamically adjust groundwater exploitation strategies. When the exploitation volume approaches or exceeds the preset carrying capacity red line, an early warning is automatically issued. The early warning information is notified to managers through the user interface, SMS or email, and optimization suggestions are provided. Based on the analysis results and early warning information provided by the interactive digital twin platform, the exploitation plan is adjusted to ensure the sustainable use of groundwater.

[0022] Example 3, as Figure 1 、 Figure 2 As shown, based on Examples 1-2, the present invention further provides a groundwater exploitation evaluation system based on the coupling of physical mechanism and neural network, which is used to implement a groundwater exploitation evaluation method based on the coupling of physical mechanism and neural network, including: Data integration and processing module, used to collect, clean, standardize and integrate meteorological data, surface water hydrological monitoring data, geological survey data, groundwater resource development and utilization data and hydrogeological parameter data from different sources to build a unified basic data set; The geological model construction module is used to build a three-dimensional geological model of heterogeneous structures using the digital twin physics engine and borehole geological data, restore the real spatial distribution characteristics of the aquifer, and improve the physical authenticity of the model; The groundwater flow simulation module is used to embed the dynamic interaction process between surface water and groundwater in the three-dimensional geological model, simulate the coupling relationship between the two, improve the water cycle simulation system, accurately reflect the water exchange between surface water and groundwater, and enhance the model's ability to simulate the dynamic changes of the regional water cycle; Uncertainty Quantification Module, which introduces graph neural networks to replace traditional numerical solvers in groundwater flow systems, embeds physical laws, and quantifies parameter uncertainties through a Bayesian neural network layer to ensure the accuracy and reliability of simulation results, improving the model's prediction accuracy and physical consistency; The interactive digital twin platform can display the exploitable amount of groundwater in real time by building an interactive digital twin platform, supporting managers to dynamically revise exploitation strategies and carrying capacity red lines based on monitoring data, thus realizing the real-time display and dynamic management of the exploitable amount of groundwater, improving the scientificity and timeliness of water resources management, and ensuring the sustainable use of groundwater.

[0023] The above description is merely a specific embodiment of the present application, but the scope of protection of the present application is not limited thereto. Any changes or substitutions that can be easily conceived by a person skilled in the art within the technical scope disclosed in this application should be included in the scope of protection of this application. Therefore, the scope of protection of this application should be based on the scope of protection of the claims.

Claims

1. A method for evaluating groundwater exploitability based on the coupling of physical mechanism and neural network, characterized by: The following steps are involved: S1. Integrate meteorological data, surface water hydrological monitoring data, geological survey data, groundwater resource development and utilization data, and hydrogeological parameter data, and pre-process them to obtain a basic data set; S2. Use the digital twin physics engine and combine it with borehole geological data to construct a three-dimensional geological model of heterogeneous structures; S3. Embed the dynamic interaction process between surface water and groundwater into the 3D geological model to simulate the coupling relationship between the two; S4. Introducing graph neural networks to replace traditional numerical solvers in groundwater flow systems, embedding Darcy's law and mass conservation equations into the loss function, and strengthening the physical consistency of the model through physical information neural networks; S5. Introduce a Bayesian neural network layer to model the probability distribution of permeability coefficient, water supply degree or water release coefficient, and quantify the uncertainty impact of input data errors on simulation results; S6. Build an interactive digital twin platform to display the amount of groundwater that can be extracted in real time.

2. The method for evaluating groundwater recoverability based on physical mechanism and neural network coupling according to claim 1 is characterized by: Said S1 specifically includes: Collect meteorological data, surface water hydrological monitoring data, geological survey data, groundwater resource development and utilization data, and hydrogeological parameter data from various sources, and compile a list of data sources; Clean the collected data to remove duplicates, anomalies, and missing values, and standardize data from different sources and formats to unify the units and formats of the data; Integrate cleaned and standardized data into a unified data platform, build a basic data set, and classify and associate data according to geographic location and time series.

3. The method for evaluating groundwater recoverability based on physical mechanism and neural network coupling according to claim 1 is characterized by: The S2 specifically includes: Prepare the digital twin physics engine, extract cleaned and standardized borehole geological data from the basic data set, import it into the digital twin physics engine, spatially locate the data, and visualize the data points in the form of a 3D point cloud in the digital twin physics engine; Using the modeling tools of the digital twin physics engine, a 3D geological model was constructed in combination with borehole geological data. A 3D grid was generated based on the scope and resolution of the study area. Different geological layers and aquifers were divided based on borehole records and lithologic distribution data. Kriging interpolation was used to generate a continuous 3D spatial distribution of aquifers. In the digital twin physics engine, interpolation parameters for the search range were set, and the permeability coefficient and lithologic parameters were interpolated to generate a 3D continuous distribution. The heterogeneous structure of the aquifer, including pores, fractures, solution gaps and lithologic changes, is depicted in the 3D geological model. At the same time, the real spatial distribution characteristics of the aquifer are restored based on geological profiles and borehole data.

4. The method for evaluating groundwater recoverability based on physical mechanism and neural network coupling according to claim 1 is characterized by: The S3 specifically includes: The spatial location and boundary data of surface water bodies are obtained from hydrological maps, remote sensing image interpretation, and geographic information system databases. These data are then imported into a three-dimensional geological model. Based on surface water monitoring data, the initial water level and flow of the surface water bodies are set. Furthermore, combined with rainfall data, the recharge sources of precipitation and surface water are defined. The dynamic interaction between surface water and groundwater is embedded in the 3D geological model. The permeability coefficient and hydraulic conductivity coefficient are set according to the borehole geological data. The recharge path and rate of surface water to groundwater, as well as the discharge process of groundwater to surface water, are defined. Darcy's law and mass conservation equation are used to simulate the coupling relationship between surface water and groundwater. A rainfall module is added to the model to simulate the recharge effect of rainfall on surface water and groundwater, and to distribute rainfall to the corresponding areas of surface water bodies and aquifers. At the same time, the impact of evaporation and transpiration on groundwater levels and surface water volume is analyzed, and the rainfall, evaporation and transpiration processes are integrated to improve the water cycle simulation system.

5. The method for evaluating groundwater recoverability based on coupling of physical mechanism and neural network according to claim 1 is characterized in that: The S4 specifically includes: The groundwater flow system is represented as a graph structure, where nodes represent key locations in the aquifer. Each node includes attributes such as coordinates, hydraulic head, lithology, and permeability. Edges represent hydraulic connections between nodes. Each edge includes attributes such as the connection node, distance, and hydraulic conductivity. A graph neural network is constructed based on this graph structure, and the network architecture is designed to adapt to the complexity of groundwater flow. The graph neural network consists of an input layer, a graph convolutional layer, a hidden layer, and an output layer. Darcy's law and the mass conservation equation are embedded in the loss function of the graph neural network. Darcy's law is used to describe the relationship between the flow rate of groundwater in the aquifer and the head gradient. The mass conservation equation is used to ensure that the inflow of water is equal to the outflow plus the storage variable. Physical information neural network technology is used to enhance the training process of the graph neural network model. The physical information neural network simulates the physical process of groundwater flow by directly embedding physical laws into the training objectives of the network.

6. The method for evaluating groundwater recoverability based on physical mechanism and neural network coupling according to claim 5 is characterized by: The training process of the graph neural network model is as follows: Embed Darcy's law and the mass conservation equation into the loss function of the graph neural network. The loss function of the graph neural network consists of two parts: data-driven loss and physical constraint loss. The data-driven loss minimizes the mean square error between the predicted head and the actual monitored head. The physical constraint loss covers Darcy's law and the mass conservation equation. Embedding Darcy's law and the mass conservation equation into the loss function yields the physical constraint loss. Using physical information neural network technology, physical laws are directly embedded in the network's training objectives. During the training process, data-driven solutions are learned while satisfying physical laws. Gradient descent is then used to optimize network parameters while satisfying both data-driven loss and physical constraint loss. Prepare training data and validation data, train the graph neural network model, and use the gradient descent method to optimize the network parameters and minimize the total loss function. In each training step, calculate the data-driven loss and physical constraint loss, update the network parameters, and then evaluate the model performance and check whether the model satisfies the laws of physics. Adjust the network architecture and hyperparameters based on the verification results, use the trained graph neural network model to predict groundwater flow, analyze the model output, and evaluate the dynamic changes of groundwater flow.

7. The method for evaluating groundwater recoverability based on physical mechanism and neural network coupling according to claim 1 is characterized by: The S5 specifically includes: A Bayesian neural network layer is introduced into the graph neural network architecture to perform probability distribution modeling on the permeability coefficient, water supply degree, or water release coefficient. The permeability coefficient, water supply degree, or water release coefficient is modeled as a combination of mean and variance. The mean represents the expected value of the parameter, and the variance represents the uncertainty of the parameter. The Bayesian layer represents the uncertainty of the parameter by introducing probability distribution. The Bayesian neural network layer is used to quantify the input data error. The Monte Carlo sampling method is used to simulate groundwater flow under different input data error scenarios and calculate the corresponding simulation results. Through multiple sampling, the distribution characteristics of the simulation results, including mean, variance and confidence interval, are analyzed to quantify the uncertainty impact of input data error on the simulation results. Analyze and evaluate the uncertainty of simulation results. By comparing simulation results under different input data error scenarios, identify the sources of uncertainty, including parameter uncertainty, data uncertainty, and model uncertainty, and evaluate the impact of uncertainty on decision support.

8. The method for evaluating groundwater recoverability based on physical mechanism and neural network coupling according to claim 7 is characterized by: The process of quantizing the input data error is: Select the Monte Carlo sampling method, set the number of sampling times M, and randomly draw samples from the defined probability distribution; For each sampling, a set of values ​​are randomly drawn from the probability distribution of the permeability coefficient, water supply degree or water release coefficient, and the groundwater flow under different input data error scenarios is simulated. Each sampling generates a set of permeability coefficient, water supply degree or water release coefficient values, runs the groundwater flow model, and calculates the corresponding simulation results, wherein the permeability coefficient, water supply degree or water release coefficient obtained for each sampling is used as the model input parameter, and the groundwater flow model is run using the model input parameters to calculate the simulation results, which are the groundwater level. Then, the simulation results of each sampling are recorded to form a data set containing M simulation results; The simulation results obtained from multiple samplings are statistically analyzed, and their distribution characteristics, including mean, variance and confidence interval, are calculated to quantify the uncertainty impact of input data errors on the simulation results.

9. The method for evaluating groundwater recoverability based on physical mechanism and neural network coupling according to claim 8, characterized in that: The S6 specifically includes: Build an interactive digital twin platform to integrate groundwater monitoring data, model prediction data, and water resource management information. At the same time, use geographic information system technology to visualize the amount of groundwater available for extraction on a map. The interactive digital twin platform receives groundwater monitoring data in real time and uses it to dynamically calibrate the groundwater flow model. When there is a deviation between the groundwater monitoring data and the model prediction, the model correction process is automatically triggered to adjust the model parameters or update the model structure. Based on real-time monitoring data and model prediction results, when the mining volume approaches or exceeds the preset carrying capacity red line, an early warning will be automatically issued and optimization suggestions will be provided. The mining plan will then be adjusted based on the analysis results and early warning information provided by the interactive digital twin platform.

10. A groundwater exploitation quantity evaluation system based on coupling of physical mechanism and neural network, used to implement the groundwater exploitation quantity evaluation method based on coupling of physical mechanism and neural network as described in any one of claims 1 to 9, characterized in that: include: Data integration and processing module, used to collect, clean, standardize and integrate meteorological data, surface water hydrological monitoring data, geological survey data, groundwater resource development and utilization data and hydrogeological parameter data from different sources to build a unified basic data set; A geological model building module is used to construct a 3D geological model of heterogeneous structures using a digital twin physics engine and borehole geological data; The groundwater flow simulation module is used to embed the dynamic interaction process between surface water and groundwater in the 3D geological model and simulate the coupling relationship between the two; Uncertainty quantification module, which is used to introduce graph neural networks to replace traditional numerical solvers in groundwater flow systems, embed physical laws, and quantify parameter uncertainties through Bayesian neural network layers; By building an interactive digital twin platform, the amount of groundwater that can be extracted can be displayed in real time, supporting managers to dynamically revise extraction strategies and carrying capacity red lines based on monitoring data.

Images